Application Of Stable Consistent Wavelength In Optimizing Near Infrared Analysis Model Transfer Of Pulpwood

Using near-infrared spectroscopy to measure holocellulose and lignin in pulpwood rapidly and non-destructively was one of the important means to improve the intelligent manufacturing level of paper industry.However,in the process of rapid detection of pulp material properties,the spectral model established on one spectrometer often cannot predict the spectrum of wood samples measured by another instrument,which greatly limited the wide application of this technology.In order to reduce the cost of re-modeling and maintenance,the model transfer technology was used to reduce the adverse effects caused by the difference among instruments,so as to achieve model sharing,which had become a key issue in the quality control of the pulping process.In this study,the near infrared spectra and the contents of holocellulose and lignin of 84 pulp wood samples in5 kinds were investigated in detail.The spectra were measured on 4 near-infrared spectrometers(3desktop machines of the same type: Lengguang 1,Lengguang 2 and Lengguang 3,a small micro near-infrared spectrometer Si-Ware).The optimization strategy of using screening wavelengths with consistent and stable signals(SWCSS)and its combined algorithm was adopted to share the near-infrared analysis models of holocellulose and lignin between different instruments,in which the Lengguang 1 was used as the master and the other three instruments as the slave.The research shown that the calibration model established by SWCSS and its combined algorithm proposed in this study significantly improved the accuracy of near infrared spectroscopy analysis of three slave machines,which layed a foundation for the popularization and application of this technology in pulp and paper industry.This paper mainly studied from the following three aspects:(1)In the process of model transfer using SWCSS algorithm,it was necessary to preprocess the spectra mesusred on master and slave before selecting representative samples by K-S algorithm for stable consistent wavelength screening.In the study,the 10 pretreatment methods such as 1 st D,2 st D,MSC,SNV and their pairwise combination was fully discussed to investigate their effects on the wavelength selection by SWCSS and then on model transfer.It was found that the model established by the wavelength selected by the SWCSS algorithm after the standardized preprocessing method had the best analysis results on the two slave samples.The RMSEP was decreased from 2.4769 to 1.2563 for one of the slave instruments and from 2.3119 to 1.2384 for another compared with the model before transfer.(2)In order to further optimize the wavelengths selected by the SWCSS method,several combined algorithms of SWCSS-UVE,SWCSS-CARS and SWCSS-SPA were proposed,that was,UVE,CARS and SPA wavelength optimization algorithms were used to reduce the adverse effects of invalid wavelengths in the SWCSS method,so as to improve the analysis ability of the model to the sampling spectra measured on two slave instrumts.The results showed that,for holocellulose,the master model the 30 wavelengths selected by the SWCSS-CARS was used to analysis the two slave sampling spectra and both RPDs were greater than 4.6,and the AIC value was 67.68,which was much smaller than 3209.83 before model transfer and 942.82 of the SWCSS algorithm.For Lignin,the models with 24 and 11 wavelengths selected by SWCSS-CARS and SWCSS-SPA methods had better analysis ability for 2 slave samples than before model transfer.Compared to using SWCSS method alone,the analysis result of two slave samples was not much different,but the AIC values decreased from 960.78 to 63.86 and 44.46,respectively,were much smaller than3198.70 before model transfer,and the model transfer efficiency was significantly improved.(3)The combined algorithms of SWCSS-DS,SWCSS-PDS and SWCSS-S/B were proposed to further improve the model transfer quality.First,the SWCSS method was used to select the wavelength with good stability and consistency to reduce the sample information error caused by the instrument or test conditions.Then,based on these selected wavelengths,direct standardization(DS),piecewise direct standardization(PDS)and slope/bias algorithm(Slope/Bias,S/B)were employed to further correct the systematic errors that still exist after SWCSS correction.The results showed that the RPDs of SWCSS-DS,SWCSS-PDS and SWCSS-S/B for the analysis of two slave samples were all greater than 3.6,which were better than those of SWCSS,DS and S/B model transfer algorithms alone,and significantly improved the analysis ability and transfer efficiency of slave samples.In the model transfer between the Lengguang 1 and the micro Si-Ware spectrometer,the analysis results of the Si-Ware measured samples by the SWCSS and SWCSSPDS methods were poor,and the RPD was less than 2.0.The analysis results of the Si-Ware measured samples by the SWCSS-DS method were better,and the RPD was greater than 3.1,which was slightly worse than the single DS algorithm.